Prabhakar Raghavan on algorithmic verification
From Prabhakar Raghavan - Can AI assist in Mathematics and Computer Science research? · · UC Berkeley EECS
“Fast and sloppy verification can lead you to a result quickly, which you then take a lot of time to verify. And just to give you a flavor of the kind of thing for the Max-4Cut, it devised a branch-and-bound strategy. And then it did some funky Python library things that I don't even understand, and came back and said, this works.”
On , Prabhakar Raghavan, Chief Technologist at Google, spoke about algorithmic verification during Prabhakar Raghavan - Can AI assist in Mathematics and Computer Science research? on UC Berkeley EECS.
About Prabhakar Raghavan
In April 2026, Prabhakar Raghavan delivered talks on the use of large language models (LLMs) in mathematics and computer science research. He described work using AlphaEvolve, an evolutionary language model from Google DeepMind, to produce new results in inapproximability and Ramsey theory, including the discovery of a gadget for the traveling salesman problem that he said would be difficult for a human to produce. Raghavan stated that he does not believe AI has yet provided a definitive capability that deepens human understanding, comparing the current state of LLMs to having "a large number of pretty smart research assistants." He also discussed the question of whether LLMs should be banned from scientific research, arguing that banning them would be "like banning microscopes from biology or telescopes from astronomy." He noted that about 20% of new computer science papers show markers of LLM-assisted writing and warned of the danger of flooding the review process with low-grade work. Raghavan and his wife Srilatha Raghavan established the Srilatha and Prabhakar Raghavan Centre for Information and Society at the Indian Institute of Technology Madras, inaugurated on April 6, 2026. In remarks at the event, Raghavan said he believes "information is empowerment and development" and framed the center's focus in the context of the Information Revolution rather than artificial intelligence specifically. He reflected on questions that arose during his leadership of information products, including how information access affects the economic trajectory of villages, democratic processes, and the adjudication of truth.